Evaluation and improvement of the Community Land Model (CLM4) in Oregon forests

Abstract

Ecosystem process models are important tools for determining the interactive effects of global change and disturbance on forest carbon dynamics. Here we evaluated and improved terrestrial carbon cycling simulated by the Community Land Model (CLM4), the land model portion of the Community Earth System Model (CESM1.0.4). Our analysis was conducted primarily in Oregon forests using FLUXNET and forest inventory data for the period 2001–2006. We go beyond prior modeling studies in the region by incorporating regional variation in physiological parameters from >100 independent field sites in the region. We also compare spatial patterns of simulated forest carbon stocks and net primary production (NPP) at 15 km resolution using data collected from federal forest inventory plots (FIA) from >3000 plots in the study region. Finally, we evaluate simulated gross primary production (GPP) with FLUXNET eddy covariance tower data at wet and dry sites in the region. We improved model estimates by making modifications to CLM4 to allow physiological parameters (e.g., foliage carbon to nitrogen ratios and specific leaf area), mortality rate, biological nitrogen fixation, and wood allocation to vary spatially by plant functional type (PFT) within an ecoregion based on field plot data in the region. Prior to modifications, default parameters resulted in underestimation of stem biomass in all forested ecoregions except the Blue Mountains and annual NPP was both over- and underestimated. After modifications, model estimates of mean NPP fell within the observed range of uncertainty in all ecoregions (two-sided P value = 0.8), and the underestimation of stem biomass was reduced. This was an improvement from the default configuration by 50% for stem biomass and 30% for NPP. At the tower sites, modeled monthly GPP fell within the observed range of uncertainty at both sites for the majority of the year, however summer GPP was underestimated at the Metolius semi-arid pine site and spring GPP was overestimated at the Campbell River mesic Douglas-fir site, indicating GPP may be an area for further improvement. The low bias in summer maximum GPP at the semi-arid site could be due to seasonal response of V_cmax to temperature and precipitation while overestimated spring values at the mesic site could be due to response of V_cmax to temperature and day length.